The gap is widening, not closing
McKinsey's 2025 State of AI report identifies a group of high performers (roughly 6% of surveyed organisations) capturing outsized returns from AI. The rest are stuck at pilot stage or reporting modest gains. The difference is not tool selection, it is operating model.
What high performers actually do differently
1. Redesign the workflow, not the task
Laggards drop copilots onto existing roles. High performers rewrite the workflow end to end, remove steps, and rebuild the handoffs so AI can carry meaningful load.
2. Embed AI in core, not periphery
High performers put AI into the processes that produce revenue and customer experience, not just internal productivity. Underwriting, pricing, service, supply chain, product development.
3. Invest in the data foundation
They build a governed knowledge and data layer so AI can access trustworthy information about the business. Without this, every use case fights the same battle from scratch.
4. Rewire governance
Executive ownership, model risk management, clear escalation paths, and disciplined measurement. AI stops being an IT project and becomes an operating discipline.
5. Upskill the operators
They build capability inside the teams that own the processes, not just in a central AI centre of excellence.
6. Measure honestly
Baseline captured, ROI defended across cost, capacity, quality, and revenue. Pilots killed quickly when the numbers do not appear.
The practical blueprint
- Pick two core processes. Not internal admin. Revenue, customer, or risk processes with clear KPIs.
- Fix the data. Build the shared knowledge layer the AI will run on.
- Redesign the workflow. Remove, standardise, then apply AI where judgement matters.
- Deploy with governance. Named owner, review cadence, audit trail, rollback plan.
- Instrument and iterate. Weekly reviews for the first quarter, then a monthly rhythm.
- Scale the pattern, not the pilot. Reuse the operating model across the next wave of processes.
Why laggards stay stuck
- Tool-first thinking: buying capability instead of redesigning work.
- Absent data foundation: every use case rebuilds context from scratch.
- Weak governance: no owner, no baseline, no consequence for failed pilots.
- Central-only capability: the people who own the processes cannot run the optimisation themselves.
Closing the gap
The gap is not permanent, but it compounds. Every quarter high performers extend their advantage because their operating model produces reusable patterns. See how this connects to the OpenGaps three-step method and the knowledge graph foundation that underpins it.